Nvidia’s Run As A De Facto Sole Source AI GPU Chip Supplier Has Come To An End

This article was originally published on 7/1/2023 and is one of the key/foundational articles being provided for reference.

In an influential blog post titled " Training LLMs with AMD MI250 GPUs and MosaicML”, MosaicML announced that LLM training software written for Nvidia (NVDA) GPUs works out of the box on Advanced Micro Devices (AMD) data center GPUs, at high-performance, and with zero code changes.

This is a major development because much of the AI software to date has been developed on Nvidia hardware and software platforms. As MosaicML notes, “Unless a new hardware platform has significant performance benefits, most users have no incentive to spend time porting their software. With new model architectures and research coming out every day, waiting for software to catch up is simply not an option. Flexibility and compatibility with existing training frameworks is a must-have for new hardware platforms.”

This dynamic has made it a challenge for new chip companies to field alternativ…